One Shot vs. Iterative: Rethinking Pruning Strategies for Model Compression